Graduate matrix map-point procs out of SkOpts.

These are implemented generically with Sk4s and don't benefit
from anything fancier than vanilla SSE/NEON.

This means there's no need to hide this code away in another
file or behind a function pointer... it's readable and we have
compile-time support for all the instructions it needs.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1872193002
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review URL: https://codereview.chromium.org/1872193002

1 files changed, 82 insertions, 7 deletions

diff --git a/src/core/SkMatrix.cpp b/src/core/SkMatrix.cpp
index f60ce85a6c..5711bd9aef 100644
--- a/src/core/SkMatrix.cpp
+++ b/src/core/SkMatrix.cpp
@@ -5,13 +5,12 @@
  * found in the LICENSE file.
  */
 
-#include "SkMatrix.h"
 #include "SkFloatBits.h"
+#include "SkMatrix.h"
+#include "SkNx.h"
+#include "SkPaint.h"
 #include "SkRSXform.h"
 #include "SkString.h"
-#include "SkNx.h"
-#include "SkOpts.h"
-
 #include <stddef.h>
 
 static void normalize_perspective(SkScalar mat[9]) {
@@ -917,11 +916,62 @@ void SkMatrix::Identity_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[
 }
 
 void SkMatrix::Trans_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_translate(m,dst,src,count);
+    SkASSERT(m.getType() <= SkMatrix::kTranslate_Mask);
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        if (count & 1) {
+            dst->fX = src->fX + tx;
+            dst->fY = src->fY + ty;
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        count >>= 1;
+        if (count & 1) {
+            (Sk4s::Load(src) + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            (Sk4s::Load(src+0) + trans4).store(dst+0);
+            (Sk4s::Load(src+2) + trans4).store(dst+2);
+            src += 4;
+            dst += 4;
+        }
+    }
 }
 
 void SkMatrix::Scale_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_scale_translate(m,dst,src,count);
+    SkASSERT(m.getType() <= (SkMatrix::kScale_Mask | SkMatrix::kTranslate_Mask));
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        SkScalar sx = m.getScaleX();
+        SkScalar sy = m.getScaleY();
+        if (count & 1) {
+            dst->fX = src->fX * sx + tx;
+            dst->fY = src->fY * sy + ty;
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        Sk4s scale4(sx, sy, sx, sy);
+        count >>= 1;
+        if (count & 1) {
+            (Sk4s::Load(src) * scale4 + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            (Sk4s::Load(src+0) * scale4 + trans4).store(dst+0);
+            (Sk4s::Load(src+2) * scale4 + trans4).store(dst+2);
+            src += 4;
+            dst += 4;
+        }
+    }
 }
 
 void SkMatrix::Persp_pts(const SkMatrix& m, SkPoint dst[],
@@ -953,7 +1003,32 @@ void SkMatrix::Persp_pts(const SkMatrix& m, SkPoint dst[],
 }
 
 void SkMatrix::Affine_vpts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_affine(m,dst,src,count);
+    SkASSERT(m.getType() != SkMatrix::kPerspective_Mask);
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        SkScalar sx = m.getScaleX();
+        SkScalar sy = m.getScaleY();
+        SkScalar kx = m.getSkewX();
+        SkScalar ky = m.getSkewY();
+        if (count & 1) {
+            dst->set(src->fX * sx + src->fY * kx + tx,
+                     src->fX * ky + src->fY * sy + ty);
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        Sk4s scale4(sx, sy, sx, sy);
+        Sk4s  skew4(kx, ky, kx, ky);    // applied to swizzle of src4
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            Sk4s src4 = Sk4s::Load(src);
+            Sk4s swz4 = SkNx_shuffle<1,0,3,2>(src4);  // y0 x0, y1 x1
+            (src4 * scale4 + swz4 * skew4 + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+    }
 }
 
 const SkMatrix::MapPtsProc SkMatrix::gMapPtsProcs[] = {